/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* vim: set ts=8 sts=2 et sw=2 tw=80: */ /* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ // Moz headers (alphabetical) #include "nsCRTGlue.h" #include "nsError.h" #include "nsIFile.h" #include "nsINIParser.h" #include "mozilla/ResultExtensions.h" #include "mozilla/URLPreloader.h" using namespace mozilla; nsresult nsINIParser::Init(nsIFile* aFile) { nsCString result; MOZ_TRY_VAR(result, URLPreloader::ReadFile(aFile)); return InitFromString(result); } static const char kNL[] = "\r\n"; static const char kEquals[] = "="; static const char kWhitespace[] = " \t"; static const char kRBracket[] = "]"; nsresult nsINIParser::InitFromString(const nsCString& aStr) { nsCString fileContents; char* buffer; if (StringHead(aStr, 3) == "\xEF\xBB\xBF") { // Someone set us up the Utf-8 BOM // This case is easy, since we assume that BOM-less // files are Utf-8 anyway. Just skip the BOM and process as usual. fileContents.Append(aStr); buffer = fileContents.BeginWriting() + 3; } else { if (StringHead(aStr, 2) == "\xFF\xFE") { // Someone set us up the Utf-16LE BOM nsDependentSubstring str(reinterpret_cast(aStr.get()), aStr.Length() / 2); AppendUTF16toUTF8(Substring(str, 1), fileContents); } else { fileContents.Append(aStr); } buffer = fileContents.BeginWriting(); } char* currSection = nullptr; // outer loop tokenizes into lines while (char* token = NS_strtok(kNL, &buffer)) { if (token[0] == '#' || token[0] == ';') { // it's a comment continue; } token = (char*)NS_strspnp(kWhitespace, token); if (!*token) { // empty line continue; } if (token[0] == '[') { // section header! ++token; currSection = token; char* rb = NS_strtok(kRBracket, &token); if (!rb || NS_strtok(kWhitespace, &token)) { // there's either an unclosed [Section or a [Section]Moretext! // we could frankly decide that this INI file is malformed right // here and stop, but we won't... keep going, looking for // a well-formed [section] to continue working with currSection = nullptr; } continue; } if (!currSection) { // If we haven't found a section header (or we found a malformed // section header), don't bother parsing this line. continue; } char* key = token; char* e = NS_strtok(kEquals, &token); if (!e || !token) { continue; } SetString(currSection, key, token); } return NS_OK; } bool nsINIParser::IsValidSection(const char* aSection) { if (aSection[0] == '\0') { return false; } const char* found = strpbrk(aSection, "\r\n[]"); return found == nullptr; } bool nsINIParser::IsValidKey(const char* aKey) { if (aKey[0] == '\0') { return false; } const char* found = strpbrk(aKey, "\r\n="); return found == nullptr; } bool nsINIParser::IsValidValue(const char* aValue) { const char* found = strpbrk(aValue, "\r\n"); return found == nullptr; } nsresult nsINIParser::GetString(const char* aSection, const char* aKey, nsACString& aResult) { if (!IsValidSection(aSection) || !IsValidKey(aKey)) { return NS_ERROR_INVALID_ARG; } INIValue* val; mSections.Get(aSection, &val); while (val) { if (strcmp(val->key, aKey) == 0) { aResult.Assign(val->value); return NS_OK; } val = val->next.get(); } return NS_ERROR_FAILURE; } nsresult nsINIParser::GetString(const char* aSection, const char* aKey, char* aResult, uint32_t aResultLen) { if (!IsValidSection(aSection) || !IsValidKey(aKey)) { return NS_ERROR_INVALID_ARG; } INIValue* val; mSections.Get(aSection, &val); while (val) { if (strcmp(val->key, aKey) == 0) { strncpy(aResult, val->value, aResultLen); aResult[aResultLen - 1] = '\0'; if (strlen(val->value) >= aResultLen) { return NS_ERROR_LOSS_OF_SIGNIFICANT_DATA; } return NS_OK; } val = val->next.get(); } return NS_ERROR_FAILURE; } nsresult nsINIParser::GetSections(INISectionCallback aCB, void* aClosure) { for (auto iter = mSections.Iter(); !iter.Done(); iter.Next()) { if (!aCB(iter.Key(), aClosure)) { break; } } return NS_OK; } nsresult nsINIParser::GetStrings(const char* aSection, INIStringCallback aCB, void* aClosure) { if (!IsValidSection(aSection)) { return NS_ERROR_INVALID_ARG; } INIValue* val; for (mSections.Get(aSection, &val); val; val = val->next.get()) { if (!aCB(val->key, val->value, aClosure)) { return NS_OK; } } return NS_OK; } nsresult nsINIParser::SetString(const char* aSection, const char* aKey, const char* aValue) { if (!IsValidSection(aSection) || !IsValidKey(aKey) || !IsValidValue(aValue)) { return NS_ERROR_INVALID_ARG; } INIValue* v; if (!mSections.Get(aSection, &v)) { v = new INIValue(aKey, aValue); mSections.Put(aSection, v); return NS_OK; } // Check whether this key has already been specified; overwrite // if so, or append if not. while (v) { if (!strcmp(aKey, v->key)) { v->SetValue(aValue); break; } if (!v->next) { v->next = MakeUnique(aKey, aValue); break; } v = v->next.get(); } NS_ASSERTION(v, "v should never be null coming out of this loop"); return NS_OK; } nsresult nsINIParser::DeleteString(const char* aSection, const char* aKey) { if (!IsValidSection(aSection) || !IsValidKey(aKey)) { return NS_ERROR_INVALID_ARG; } INIValue* val; if (!mSections.Get(aSection, &val)) { return NS_ERROR_FAILURE; } // Special case the first result if (strcmp(val->key, aKey) == 0) { if (!val->next) { mSections.Remove(aSection); } else { mSections.Put(aSection, val->next.release()); delete val; } return NS_OK; } while (val->next) { if (strcmp(val->next->key, aKey) == 0) { val->next = std::move(val->next->next); return NS_OK; } val = val->next.get(); } return NS_ERROR_FAILURE; } nsresult nsINIParser::DeleteSection(const char* aSection) { if (!IsValidSection(aSection)) { return NS_ERROR_INVALID_ARG; } if (!mSections.Remove(aSection)) { return NS_ERROR_FAILURE; } return NS_OK; } nsresult nsINIParser::RenameSection(const char* aSection, const char* aNewName) { if (!IsValidSection(aSection) || !IsValidSection(aNewName)) { return NS_ERROR_INVALID_ARG; } if (mSections.Get(aNewName, nullptr)) { return NS_ERROR_ILLEGAL_VALUE; } nsAutoPtr val; if (mSections.Remove(aSection, &val)) { mSections.Put(aNewName, val.forget()); } else { return NS_ERROR_FAILURE; } return NS_OK; } nsresult nsINIParser::WriteToFile(nsIFile* aFile) { nsCString buffer; for (auto iter = mSections.Iter(); !iter.Done(); iter.Next()) { buffer.AppendPrintf("[%s]\n", iter.Key()); INIValue* val = iter.Data(); while (val) { buffer.AppendPrintf("%s=%s\n", val->key, val->value); val = val->next.get(); } buffer.AppendLiteral("\n"); } FILE* writeFile; nsresult rv = aFile->OpenANSIFileDesc("w", &writeFile); NS_ENSURE_SUCCESS(rv, rv); unsigned int length = buffer.Length(); if (fwrite(buffer.get(), sizeof(char), length, writeFile) != length) { fclose(writeFile); return NS_ERROR_UNEXPECTED; } fclose(writeFile); return NS_OK; }